Department of Economics Working Paper Series From Policy Preferences to Partisan Support: A Quantitative Assessment of Political Culture in South Dakota Dakota Cannabis use and suicidal ideation Russell H. Hillberry, William D. Anderson August 2012 Research Paper Number 1156 ISSN: 0819 2642 ISBN: 978 0 7340 4507 2 Department of Economics The University of Melbourne Parkville VIC 3010 www.economics.unimelb.edu.au From Policy Preferences to Partisan Support: A Quantitative Assessment of Political Culture in South Dakota Russell H. Hillberry∗ William D. Anderson† University of Melbourne University of South Dakota August 2012 Abstract This study uses cross-county variation in support for 46 ballot measures to identify political subcultures in South Dakota and to study them. A hierarchical clustering method applied to county-level election returns allows the identification of subcultures at various levels of granularity. We choose a threshold that suggests seven subcultures as a useful summary of the data. While the allocation procedure employs only election returns as an input, the identified subcultures match observable regularities in demographics and geography. Subsequent factor analysis of election returns from the ballot measures reveals a multi-dimensional policy space. By contrast, a similar analysis of support for political candidates reveals a single partisan spectrum as a dominant feature of the data. A county’s location in the revealed policy space well explains its location along this partisan spectrum. The link between policy and partisan preferences is robust to the inclusion of a wide variety of additional control measures. Keywords: Constraint, political culture, partisanship, efficient classification ∗Corresponding author. Department of Economics, University of Melbourne, Victoria 3010, Australia email: [email protected]. †Department of Political Science, University of South Dakota, 414 East Clark Street Vermillion, South Dakota 57069, USA. 1 1 Introduction Legislative voting patterns exhibit a high degree of structure; a one-dimensional spectrum explains most of the variation in legislative votes.1 Studies based on polling of individual voters, however, show that the members of the public generally exhibit minimal ‘constraint’ in their policy preferences; a voter’s position on one issue generally does not predict that voter’s position on another.2 Why and how ideological/partisan structure in legislative politics emerges from the unstructured political preferences of the citizenry is a question of considerable importance to students of politics.3 We investigate intermediate stages of this process, identifying regularities in voters’ (aggregated, county-level) responses to specific pieces of proposed legislation (i.e. ballot measures), and corresponding levels of support for partisan candidates for political office. The literature on political culture (Elazar 1984) provides a useful framework for organizing our thinking on these matters. This literature argues that ethnic, religious and other characteristics of locally dominant demographic groups help to shape state and/or local political institutions and outcomes. In contrast to the literature on constraint, these analyses operate under the maintained hypothesis that voting populations’ aggregated political behavior are consistent with underlying world views that are, to some degree, coherent and identifiable. A further presumption is that these world views are shaped by demographic and other criteria. This paper introduces a new method for identifying political cultures, a hierarchical cluster analysis of election returns data. We apply this method to county-level election returns from 46 recent ballot measures in South Dakota.4 The method allows political subcultures to be 1See Poole and Rosenthal (1997) among others. 2Converse (1964) introduces the concept of constraint in belief systems and provides the initial evidence that it is lacking in the political attitudes of the mass public. 3Feldman (2003, p.478), quoted in Linzer (2006), argues that unstructured political attitudes among the public are problematic for political theories of democracy. The lack of structure among voters, he argues, makes communication between elected officials and voters difficult, and undermines concepts of democratic representation. 4Election returns offer somewhat different information than do polling data. Ballot measures will take the force of law if enacted, responses are completely anonymous, and the outcomes follow campaigns that might shape opinions on the issues in question. One might better expect responses to reflect the considered opinions of the voters. The disadvantages of ballot measure data are 1) that they are aggregated (so we can 2 identified at various levels of granularity. At each level of granularity, the identified county groupings are plausibly linked to local demography, even though the classification procedure relies only on the political behavior of voters, not the demographic characteristics of the counties in which they reside. The results provide support to a key argument in the political culture literature: that local demographics drive political outcomes.5 The same data can be used to sketch the ideological landscape of South Dakota. Like authors investigating regularities in legislative voting behavior, we apply formal dimension- reduction techniques to our data on electoral responses. Factor analysis over the 46 referenda and initiated measures considered by South Dakota’s voters over a 12-year period produces a seven-dimensional space, with only three dimensions explaining 70 percent of the overall cross-county variation in election returns. The factor analysis highlights the issues of primary division between the identified political subcultures. There appear to be consistent sources of division over related issues. These results further indicate more cohesion in electoral outcomes than the constraint literature would suggest, but still considerably less than is evident in studies of the legislative arena.6 In order to put the multidimensional nature of votes on the ballot measures in context we next turn to an analysis of electoral support for partisan candidates for state-wide political office. These elections include races for very different offices, with very different responsibilities (U.S. President, seats in the U.S. House and Senate, and South Dakota’s Governorship). A factor analysis of county-level election returns from 1996-2006 returns a dominant single factor not investigate regularities among individuals), 2) they may not cover as broad an area of the policy space as polls that are designed to investigate constraint. The large number of recent ballot measures faced by South Dakota voters offer a partial solution to the latter problem. 5The data do not allow us to distinguish between two possible reasons that demographics matter. First, it may be that demographic sub-populations have particular policy preferences, and the views of larger sub-populations dominate aggregate outcomes. Second, it may be that dominant populations affect world views of their neighbours. A claim that the first of these positions was the reason for our findings would be subject to critique based on the ecological fallacy. We are not making such claims, and we take arguments about the ecological fallacy to be tangential to our purpose. The behavior of aggregates is important in its own right, as it is aggregated voter behavior that determines political outcomes in a democratic system. In this, we concur with Snyder (1996; p. 464). 6Banducci (1998) reaches similar conclusions from a study of individual ballots from a single Oregon county in 1990. In her study, individual voters demonstrated a moderate degree of constraint. Three dimensions capture most of the variation in support for the eight ballot measures. 3 that explains more than 80 percent of the cross-county variation in support for the candidates. This is a partisan spectrum, and a robust one, a point made clear by the predictive power of county-level factor scores for outcomes in the out-of-sample state-wide elections of 2008. The tightness of these links suggest that high levels of partisan coherence emerge when voters have partisan labels available to them.7 Next we link our identified political subcultures to levels of partisan support for the candidates in state-wide races. We regress levels of partisan support on dummy variables representing the seven nominated subcultures. This regression reveals high levels of explanatory power. We take this as joint evidence that 1) the classification procedure has considerable merit, and 2) political subcultures as we have measured them explain a substantial portion of the variation in support for partisan candidates. Finally, we turn to a more nuanced regression of county-level factor scores from the races for political office on the factor scores from our analysis of ballot measures. These regressions serve two purposes. First, they establish an even tighter link between counties’ positions on the ballot measures and their partisan preferences in races for public office. Second, the regression coefficients offer an analytic guide to the way in which the complex ideological space suggested by patterns of support for the ballot measures is reduced to a single, dominant partisan spectrum. We find that a single factor from our analysis of the ballot measures explains more than half the cross-county variation along the partisan spectrum.8 Moreover, the seven factors that efficiently summarize